Conclusiones

For this study, close out performance data were obtained for groups of cattle that remained together as a single lot from arrival until slaughter. Lots of cattle were weighed together on trucks, at the start and end of the finishing feeding period. The fecal samples (pool of 4 as

described above) were collected from a total of 292 pens corresponding to 90 lots of cattle where DMI, ADG, and G:F could be obtained. Of these lots, 70 contained steers, and 20 contained

heifers. Measures of interest (i.e. fecal composition, digestibility, grain percent, F:C, grain processing) were averaged if samples were collected from the same lot more than once. The frequency of fecal collections within a feeding period for each lot was recorded.

Close out DMI was estimated by the amount of feed offered to a lot (adding the total feed offered to each pen making up a lot) over the full feeding period, divided by the number of days on feed and the number of cattle in the lot. Close out ADG and G:F were calculated assuming 4% shrink. Close out data were calculated by subtracting average initial weight from the average final shrunk body weight with dead animals removed. This difference was divided by the number of cattle, and the days on feed for each pen, as described by Gaylean et al. (2010). The pen G:F was calculated as ADG divided by DMI.

7.2.6 Statistical Analysis

The Mixed procedure in SAS v. 9.1. (SAS Inst. Inc., Cary, NC, USA) was used to determine the differences in management strategies (grain percent, F:C, processing index), in NIRS

predicted fecal composition and digestibility, and performance, observed among the six feedlots.

Pen was used as the experimental unit, but the number of repetitions varied because of frequent shipping of cattle and because the distribution of cattle within pens varied among feedlots and with month. A compound symmetry structure was used to model repeated measures on individual pens sampled on more than one day as it exhibited best fit for convergence. In addition to fecal starch and fecal NDF, aTTD of DM, OM, and starch were selected as key interests for digestibility based on previous investigations (Chapter 5; Owens et al. 2016). The aTTD of GE (%) was also selected because of its relationship to net energy of gain (Chapter 5).

The model used was Yij = β0ij + β1feedlotij + β2dayij + j + εij, where β0 was the average

y-intercept or overall mean for management factors, fecal chemical composition, digestibility, BW, DMI, ADG, or G:F when all terms in the equation are equal to 0, i was the individual

observation of day within pen, j was feedlot, j was the residual associated with feedlot and εij

was the residual error associated with each observation. Least square means of the treatments were separated using PDIFF statement, with significance declared at P < 0.05.

The Mixed procedure was then used to determine the effect of grain type, processing method, and sex on NIRS predicted fecal composition, digestibility and growth performance, adjusting for differences among feedlots as a random intercept. Pen was used as the experimental

unit, and a compound symmetry structure was used to model repeated measures on individual pens. Interactions were not examined in this observational study because of the large number of variables examined, unequal numbers of pens per lot and feedlot, and missing data. The model used was Yij = β0ij + β1grain typeij + β2processing methodij + β3sexij + β4seasonij + β5dayij+ j + εij, where β0 was the average y-intercept or overall mean for fecal chemical composition, digestibility, DMI, ADG, or G:F when all terms in the equation are equal to 0, i was the individual observation of day within pen, j was feedlot, j was the residual associated with feedlot and εij was the residual error associated with each observation. Least square means of the treatments were separated using PDIFF statement, with significance declared at P < 0.05.

A third series of regression models were used to examine the bivariate associations between each of fecal nutrients and digestibility measures, and each measure of performance, including DMI, ADG, and G:F at each lot. Lot of cattle was considered the experimental unit for this analysis, and if multiple fecal collections occurred within the same lot of cattle, fecal nutrients and digestibility measures were averaged before inclusion in the regression. Manual backward stepwise regression was used to generate multivariable equations to predict DMI, ADG, or G:F for each lot of cattle using information on fecal nutrients and NIRS predicted digestibility, with a p-value for bivariate association of < 0.20 suggesting a potential association. The mixed linear regression model included a random effect for feedlot and fixed effects for grain type, grain percent, processing method, processing index, F:C, sex, season, and average BW of the cattle at the time of sampling.

The correlations between all pairs of variables that were significantly associated with an outcome of interest were assessed using PROC CORR. Where variables were highly correlated (r

≥ 0.90, P < 0.01), the variable from the pair with the greatest P value was removed before consideration in building the final multivariable model. The linearity assumption was examined for each of the continuous measures of fecal nutrient concentrations and digestibilities in these models. Differences were declared significant at P < 0.05, and trends at P < 0.10 for all models.

The concordance correlation coefficient (CCC) (rho; Lin 1989; 2000) was used to calculate the concordance between observed and predicted DMI, ADG, and G:F. The CCC reflects the agreement between two sets of results with a value of 1 indicating perfect agreement between the results. Differences were declared significant at P < 0.05, and trends at P < 0.10 for all models.

7.3 Results and Discussion 7.3.1 Accuracy of NIRS Predictions of Fecal Samples

A pre-requisite for successful NIRS of feces is that the spectral variability of samples to be analyzed is encompassed by the calibration database (Landau et al. 2015), a requirement satisfied by the calibrations in the present study. Calibrations should not be static, but rather continuously evolve as new datasets become available. To expand previous calibrations, data selection

methods such as random selection, manual selection, and discriminant analysis by wavelength selection or PCA can be used (Westerhaus et al. 2004). These methods recognize outlier samples that are identified using H (or Mahalanhobis) distances, Hotelling’s T² distribution (WinISI 1.50, Infrasoft International, Silver Spring, MD), Global and Neighborhood distances (Ucal, Unity Scientific, 2010), and T statistics (Ucal), depending on the type of software employed. In the current study, we used manual selection to expand the original calibration by including a subset of commercial feedlot samples that were selected based on NIRS predicted constituent

concentrations. We then examined ND in the validation set, and added samples with ND > 0.60.

Prior to adding the subset of samples into the initial calibration for chemical composition, the coefficients of determination of validation (R²val) was 0.94 for fecal starch, 0.70 ≤ R²val ≤ 0.80 for fecal OM, N, NDF, ADL, and R²val = 0.25 for fecal ADF, and EE (Chapter 3). We did not perform a second validation or additional reference analysis after initial calibration expansion, but it is likely that this step improved predictability. The coefficients of determination of cross-validation increased, or remained high (except for EE), and the standard errors of cross

validation did not dramatically change (Chapter 3). Obviously, total tract digestibility could not be measured in commercial feedlot cattle. Consequently, all samples in the calibration set used to predict digestibility were collected from cattle housed indoors during metabolism experiments (Chapter 3). As a result, none of the commercial feedlot samples could be used to expand the digestibility calibration set.

Discriminant analysis (PCA) was used to visualize and compare spectral populations as a whole using the Hotelling’s T² distribution. Samples outside of the defined Hotelling’s T² ellipses were identified as outliers from the population to a confidence level of 95%. Principal component analysis demonstrated substantial overlap along PC 1 and PC 2 between the calibration set for fecal chemical composition and the samples collected from the commercial

A

B

Figure 7.1. Principal component graph displaying (A) the spectra of fecal samples collected from commercial feedlots over a 1 year period in relation to the fecal samples in the NIRS calibration set for chemical composition along the first two principal components, and (B) the spectra of fecal samples collected from six commercial feedlots over a 1 year period in relation to the fecal samples in the NIRS calibration set to estimate digestibility along the first two principal components. Samples outside of the defined Hotelling’s T² ellipses are considered outliers from the population to a confidence level of 95%.

feedlots (Figure 7.1A). Principal components 1 and 2 explained 64% of the variation in the calibration samples, and 41% in the feedlot samples. The only samples from the feedlot dataset that were considered outliers according to the Hotelling’s T² distribution were samples collected in January. There was much less overlap between the calibration sets for digestibility and those collected from the commercial feedlots, and two separate populations were evident (Figure 7.1B).

The first two PCs still explained a large proportion of the spectral variability in both datasets (67% for the calibration samples, and 44% for the feedlot samples) and feedlot samples collected in January were again identified as outliers, as well as a few samples collected in November and December. It is possible that the cold temperatures that cattle were exposed to during this period may have altered the composition of feces as compared to those used in the development of digestibility equations which were all collected indoors at room temperature. This is consistent with previous work by Coates and Dixon (2012) where calibrations for predicting diet

digestibility in ruminants were developed using samples collected over 10 years using various sampling methods. They demonstrated that experimental site and sampling method often had important effects on calibration statistics and performance. Landau et al. (2015) also developed NIRS calibrations of feces for predicting dietary composition in beef cattle in east Mediterranean rangelands, and found that seasonal trends in pasture quality and responses to management practices impacted estimates. This would explain the differences shown in the PCA graphs between spectra collected from metabolism studies in Chapter 3, and those collected from the commercial feedlots in the current study.

7.3.2 Characterization of Fecal Samples

To our knowledge, this study is the first to examine the extent to which NIRS can be used to predict the composition of feces, diet digestibility and growth performance of commercial feedlot cattle. The average fecal DM, and NIRS predicted fecal composition and digestibility estimates (Table 2) were comparable to estimates from a previous study (Chapter 3). The previous datasets were compiled from fecal samples representing over 60 diets from both research feedlot and metabolism studies. Although none of the average fecal concentrations reported in the commercial feedlot samples appeared unusual, the largest discrepancy identified occurred with fecal NDF, which was on average 3% lower than previous estimates from research

feedlot studies. This is not likely due to an error with NIRS as the average fecal NDF in a subset of these samples that were analyzed using wet chemistry was also lower than reported in

previous studies (46.0 ± 8.02%; Chapter 4). As we did not analyze the diets that were fed to the commercial feedlot cattle, it was impossible to determine if the lower fecal NDF was a result of lower NDF intake.

The commercial feedlot fecal samples possessed individual samples with unusually low and high estimates of starch concentration. For example, 5 fecal samples had predicted starch

concentrations of less than 1% starch. These samples came from separate pens collected within the same month from two feedlots that were feeding temper-rolled barley. The standard error in the calibration for starch was 1.67%, which indicates an error of ±1.67% fecal starch DM in 68%

of samples (based on a normal Gaussian distribution), and ±3.34% in 95% of the samples. In a previous study, NIRS was used to predict starch in fecal samples collected from backgrounding cattle, and if starch levels in feces were low (≤ 2%), the average C.V. for duplicate samples were much higher than wet chemistry estimates (0.33 versus 0.055; Chapter 4). In the finishing period, fecal starch concentrations were higher and the average C.V. for NIRS analysis were only slightly higher than wet chemistry (0.067 versus 0.053; Chapter 4). All of the fecal samples with starch values > 16% originated from a single feedlot that exhibited the highest processing index.

The nutrient and GE digestibility values in the 6 commercial feedlots were higher than those reported in previous studies (Chapter 3) and unrealistically high (up to 100%) estimates in aTTD of GE were observed (Table 2). The reason for this high variability in prediction in field as compared to our previous research studies is unknown. In commercial feedlots, competition among cattle for feed can be intense if meal delivery is delayed or if bunk space is limiting.

Competition is known to increase consumption rates and reduce eating time (Olofsson 1999), factors that can affect digestibility. Furthermore, because dominant and subordinate cattle are penned together, eating behaviours of these cattle differ from the individually housed cattle (Striklin and Gonyou 1981) that were used in metabolism studies to develop the NIRS calibrations to predict digestibility. Strengthening of the calibration equations through the addition of data from a broader range of digestibility studies may help improve these estimates.

Since the PCA graphs show different populations, it is also possible that the NIRS

predictions lacked accuracy, and were over-predicting actual values. Despite the ability of NIRS

Table 7.2. Simple statistics of fecal DM and NIRS predicted fecal composition and apparent total tract digestibility derived from pen composite fecal samples collected from feedlot cattle fed in six commercial feedlots (n=282 pens) over a 1 year period

Item Measurement¹

Mean±SD C.V. (%) Min Max

Fecal composition,

% of fecal DM

Dry matter 18.9±2.31 12.2 13.6 27.0

Organic matter 83.8±3.58 4.27 67.5 89.4

Starch 7.0±3.87 55.3 0.00 25.1

Nitrogen 2.36±0.215 9.11 1.83 3.00

NDF 50.4±4.78 9.48 33.2 60.5

ADF 29.3±3.18 10.8 19.9 36.2

ADL 5.41±0.939 17.3 3.29 8.03

Ether extract 1.50±0.183 12.2 0.849 1.912

Digestibility, % of intake

Dry matter 82.0±3.28 4.00 70.4 89.2

Organic matter 81.6±3.53 4.32 70.0 89.3

Starch 95.0±2.14 2.25 86.2 99.5

Crude protein 77.6±3.04 3.92 68.4 85.0

NDF 62.0±3.56 5.74 53.5 73.5

ADF 46.9±5.10 10.9 27.8 63.6

Gross energy 88.0±5.19 5.90 74.6 100

1SD = standard deviation; C.V. = coefficient of variation (SD/mean)

to be less predictive of digestibility as compared to the chemical composition of feces, the current method has value in terms of its potential to estimate diet digestibility in group-penned feedlot cattle without using markers or total collection techniques. Chapter 3 reported that even though predictions of certain digestibility coefficients using NIRS were not identical to estimates obtained from total collection, expected changes in digestibility in response to dietary changes could be predicted, illustrating the merit of this approach.

Variation in fecal chemical composition and diet digestibility among different individual cattle is important to address when attempting to predict outcomes from a population of cattle without sampling all of them. Consistent with Chapter 5, where fecal starch values within a pen varied substantially, a high degree of variation was also observed in fecal starch (C.V. = 55%) in the commercial feedlot samples. In contrast, all other fecal constituents were reported to have C.V.

of 12% or less (Table 2). The greatest variability in digestibility was observed for aTTD of ADF, an outcome that was attributed to the relatively poor linearity and accuracy of predicting this fecal constituent (Chapter 3).

7.3.3 Differences between Management Practices, Fecal and Performance Measures Beef cattle performance is dependent on many factors including the initial BW, age, sex, body condition, health record (Reinhardt et al. 2009; McMeniman et al. 2010; Galyean et al.

2010), animal behaviour, hierarchical behaviour, diet, inclusion of feed additives and anabolic agents, as well as season and temperature (NASEM, 2016). Our sample size was limited (6 feedlots and 4 pens per feedlot) and therefore we did not have sufficient statistical power

investigate interactions among production parameters. The factors that we could account for such as grain type, grain processing method, sex, season, and day were adjusted for in the analysis.

Other factors such as the effects of grain inclusion rate, F:C ratio, and diet ingredients were reported as well. For example, grain percent was quite low for feedlots 1 (69.2 ± 5.08) and 6 (64.7±5.10) in relation to others (all ≥ 77.6%), with feedlot 1 also having the highest F:C ratio (0.16 versus ≤ 0.11 for the remaining five feedlots; Table 3). Without any additional information, we could expect that cattle in this feedlot would have the poorest performance because of the lower energy density of these diets. However, this was not the case as the DMI was the greatest for feedlot 1, resulting in increased energy intake. Also, the degree of grain processing and method utilized must also be considered as the least vigorous grain processing occurred in

feedlot 5 (84.3±8.27; Table 3). This likely explains why cattle in feedlot 5 exhibited the lowest (P ≤ 0.01) ADG (1.19 kg BW/d) and G:F (0.110), the highest (P ≤ 0.01) fecal starch (11.6% of fecal DM) and the lowest (P ≤ 0.01) fecal NDF (45.7% of fecal DM) concentration of the feedlots examined (Table 3).

To clarify other discrepancies, despite feedlot 1 having the lowest percentage of grain in the diet, and greatest F:C, performance was not compromised. In times of poor weather or low DMI as a result of rapid diet changes, feedlot 1 fed cattle a finishing diet with lower levels of grain for long periods of time. This occurred over the course of several sampling times, and would have resulted in the lower grain percentage and higher F:C ratio reported. Feedlot 6 always temper-rolled grain and included potatoes in the diet, both of which would increase the energy content of the diet. Feedlot 5 was the most obvious outlier, where processing index and fecal starch

concentration were exceptionally high, and digestibility coefficients low, factors that likely contributed to the reduced growth performance of cattle at this location.

Many western Canadian feedlots use barley or wheat that is processed by dry or temper rolling (McAllister et al. 2011). When grain is dry-rolled, the degree of fracturing of the grain kernel is dependent on the adjustable distance between two rollers. This processing method is sufficient for barley and wheat, but can result in greater variability in particle size including the generation of fine particles and dust (Wang et al. 2003; McAllister et al. 2011). Fines (particles less than 1 mm diameter) have been shown to increase the occurrence of digestive dysfunction due to starch digestion being too rapid acid accumulating in the rumen (Mathison, 2000). Temper rolling involves application of water for 8 to 24 h prior to rolling, enabling more control over the degree of fracture of the kernel and reducing the generation of fine particles. Industry standards for processing index of dry rolled barley grain ranges between 65 and 82% (Yang et al. 2000), whereas temper rolled barley ranges from 70 to 95% (Beauchemin et al. 2001). Over the 1 yr period, only feedlot 5 exceeded the upper range of the recommended PI for dry rolling.

Cereal grain processing acts to increase ruminal digestion of all nutrients, especially starch, by first disrupting the pericarp (and hull in hulled grains) and secondly by reducing the grain particle size, accelerating microbial colonization and fermentation (McAllister et al. 2011). If grain is processed more vigorously, measured as a lower processing index, exposure of the starch to microbes will increase. Our results show that compared to temper rolled grain, dry rolled grain was processed less vigorously (75.4 ± 9.33 versus 66.0 ± 4.08). However, temper rolled grain

Table 7.3. Differences in NIRS predicted fecal starch and NDF, digestibility of DM and OM, performance, and management strategies in six commercial feedlots over a 1 year period (n=282)

168

Table 7.4. Differences in NIRS predicted fecal starch and NDF, digestibility of DM, OM, performance, and management strategies for dry rolled grain versus temper rolled, barley versus barley and wheat diets, and heifers versus steers, (n=282)

Item¹

Management Heifer (n=53) Steer (n=228) P value

Grain % 76.5 ± 9.47 76.0 ± 10.48

1F:C = forage to concentrate ratio; PI = processing index; B = barley; BW = barley and wheat.

2 A total of 292 fecal samples are represented in the table. Data is missing for F:C (n = 269), BW (n = 242), and grain percent (n = 272). Numbers in brackets represent the standard error of the mean. Diets containing potatoes as a source of starch were only included barley and temper rolled diets. Different letters across each row represent significant differences P < 0.05.

has a higher moisture content (17 to 25%) and greater malleability, allowing for less severe processing. Therefore, the processing indices for the two methods should not be directly

has a higher moisture content (17 to 25%) and greater malleability, allowing for less severe processing. Therefore, the processing indices for the two methods should not be directly